WandB integration and confusion matrix

[CyrusAddy]

I am performing training using YOLOv8, YOLOv9, YOLOv10, and YOLOv11 on wandb. However, I am not able to get the confusion matrix printed out as part of the loggings on wandb. Any help on how to do that?

[UltralyticsAssistant]

👋 Hello @CyrusAddy, thanks for opening this discussion and for using Ultralytics 🚀! This is an automated response to help you get quick guidance while an Ultralytics engineer reviews your post and will assist you soon.

For general usage, we recommend visiting the Docs where you can find many Python and CLI examples and answers to common questions:

• Docs: https://docs.ultralytics.com/
• Python examples: https://docs.ultralytics.com/usage/python/
• CLI examples: https://docs.ultralytics.com/usage/cli/

If this is a 🐛 Bug Report (e.g., expected confusion matrix not appearing in Weights & Biases logs), please share a minimum reproducible example to help us debug it:

• MRE guide: https://docs.ultralytics.com/help/minimum-reproducible-example/

If this is a custom training ❓ Question about configuration or integrations, please include:

• Exact Ultralytics version and model used (YOLOv8/YOLOv9/YOLOv10/YOLO11)
• OS, Python, and PyTorch versions
• W&B settings/config, relevant CLI/Python commands, and training logs
• Sample dataset images and any related screenshots
• Confirmation that you’ve reviewed our Tips for Best Training Results: https://docs.ultralytics.com/guides/model-training-tips/

Join the Ultralytics community where it suits you best:

• Discord for real-time chat 🎧: https://discord.com/invite/ultralytics
• Discourse for in-depth threads: https://community.ultralytics.com/
• Subreddit for community sharing: https://www.reddit.com/r/Ultralytics/

Upgrade

Please upgrade to the latest ultralytics package with all requirements in a Python>=3.8 environment with PyTorch>=1.8 to ensure the issue isn’t already resolved:

pip install -U ultralytics

• Requirements: https://github.com/ultralytics/ultralytics/blob/main/pyproject.toml

Environments

YOLO can be run in the following verified environments (with dependencies like CUDA/CUDNN, Python, and PyTorch preinstalled):

• Notebooks with free GPU:
• Google Cloud Deep Learning VM. Guide: https://docs.ultralytics.com/yolov5/environments/google_cloud_quickstart_tutorial/
• Amazon Deep Learning AMI. Guide: https://docs.ultralytics.com/yolov5/environments/aws_quickstart_tutorial/
• Docker Image. Guide: https://docs.ultralytics.com/yolov5/environments/docker_image_quickstart_tutorial/

Status

If this badge is green, all Ultralytics CI tests are passing. CI tests verify correct operation of all YOLO Modes and Tasks on macOS, Windows, and Ubuntu every 24 hours and on every commit.

Once you share the requested details (including an MRE if it’s a bug), we’ll be able to assist you more effectively. Thanks for your patience and for trying YOLOv8, YOLOv9, YOLOv10, and YOLO11 with W&B integration 🙌

[glenn-jocher]

Hi @CyrusAddy — W&B logs the confusion matrix as an image when validation plots are enabled; enable W&B and plots, then train or run a val pass to push it:

yolo settings wandb=True
yolo train model=yolo11n.pt data=your.yaml plots=True

or for an existing model

yolo val model=path/to/best.pt data=your.yaml plots=True

You’ll see confusion_matrix_normalized under Images in the run; this is produced by the confusion matrix plot and logged via the W&B callback that uploads validator.plots at epoch/end of training. Details are in the YOLO + Weights & Biases integration guide and the W&B callback reference.

[CyrusAddy]

@glenn-jocher thank you that worked.

[CyrusAddy]

@glenn-jocher I need the CLI command to save predictions json file. Any help?

[glenn-jocher]

Use val with save_json to write COCO-format predictions to your run directory:

yolo val model=path/to/best.pt data=your.yaml save_json=True

This will create predictions.json in the run folder; more on validation outputs in Model Validation with Ultralytics YOLO.